fix(core/transport): treat ENOBUFS as a transient drop, not a fatal error

WiFi drivers (e.g. ath11k on the Steam Deck) return ENOBUFS — not
EAGAIN/EWOULDBLOCK — when the tx queue is momentarily full. Rust maps
ENOBUFS to ErrorKind::Uncategorized, so `is_transient_io` (which only
matched WouldBlock/ConnRefused/ConnReset) treated it as a real error and
tore the whole stream down on a single transient burst.

This presented as a vicious Heisenbug on the Deck: the native host
streamed flawlessly on loopback and under a debugger (anything slow
enough not to fill the small ~416 KB wlan0 buffer), but died at full rate
cross-machine over WiFi — flaky hang-or-SIGKILL because tx-queue-full is
probabilistic. Diagnosed live via a forced core dump (gdb on the hung
core): the data-plane thread had bailed on a fatal send error.

Treat ENOBUFS (and asynchronous network-path blips ENETUNREACH /
EHOSTUNREACH / ENETDOWN / EHOSTDOWN) as a lossy drop like WouldBlock —
FEC + the next frame recover. Validated: 6/6 back-to-back cross-machine
streams over the Deck's WiFi, host stable, p50 ~4.4 ms (one run dropped
4/300 frames *gracefully*, 0 mismatched — the fix working as intended).

Also surface a data-plane bind/hole-punch failure directly in punktfunk1
(it was previously only reported after teardown, which a stall could
swallow entirely).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-06-20 14:49:59 +00:00
parent 333f66b45b
commit 0f7f1be3c3
2 changed files with 37 additions and 4 deletions
+27 -1
View File
@@ -26,9 +26,35 @@ const RECV_BUF: usize = MAX_DATAGRAM_BYTES + 1;
/// so erroring out here kills a stream that the very next packet would resume. If the peer is
/// genuinely gone, the QUIC control plane times out and ends the session cleanly instead. (This is
/// the classic connected-UDP "ICMP errors are advisory" rule, doubly true with hole-punching.)
/// - `ENOBUFS`: a WiFi/wlan driver (e.g. `ath11k` on the Steam Deck) returns this — NOT `EAGAIN`/
/// `WouldBlock` — when its tx queue is momentarily full. Rust maps `ENOBUFS` to
/// `ErrorKind::Uncategorized`, so the `WouldBlock` arm misses it; without this a transient
/// tx-queue burst tears the whole stream down (observed live: the host streamed flawlessly on
/// loopback / under a debugger — anything slow enough not to fill the small wlan0 buffer — but
/// died at full rate over WiFi). Same lossy-drop contract as `WouldBlock`; FEC + the next frame
/// recover. Asynchronous network-path blips (`ENETUNREACH`/`EHOSTUNREACH`/`ENETDOWN`/`EHOSTDOWN`)
/// are droppable for the same reason a stale ICMP is.
fn is_transient_io(e: &std::io::Error) -> bool {
use std::io::ErrorKind::{ConnectionRefused, ConnectionReset, WouldBlock};
matches!(e.kind(), WouldBlock | ConnectionRefused | ConnectionReset)
if matches!(e.kind(), WouldBlock | ConnectionRefused | ConnectionReset) {
return true;
}
// `ENOBUFS` & friends have no stable `ErrorKind`, so match the raw errno (unix only).
#[cfg(unix)]
{
matches!(
e.raw_os_error(),
Some(libc::ENOBUFS)
| Some(libc::ENETUNREACH)
| Some(libc::EHOSTUNREACH)
| Some(libc::ENETDOWN)
| Some(libc::EHOSTDOWN)
)
}
#[cfg(not(unix))]
{
false
}
}
/// Build one `mmsghdr` per `iovec` (each a single-buffer message) for `sendmmsg`/`recvmmsg`. Shared
+10 -3
View File
@@ -837,12 +837,19 @@ async fn serve_session(
// can be on different subnets; control + side planes ride the client-initiated QUIC, but
// the raw video UDP needs the client to open the path first). Falls back to the
// client-reported address for clients that don't punch (flat-LAN, unchanged).
let (transport, punched) = UdpTransport::connect_via_punch(
let (transport, punched) = match UdpTransport::connect_via_punch(
&format!("0.0.0.0:{udp_port}"),
&client_udp.to_string(),
std::time::Duration::from_millis(2500),
)
.context("bind data plane")?;
) {
Ok(v) => v,
Err(e) => {
// Surface the failure here directly: a data-plane bind error would otherwise be
// reported only after teardown (and a teardown stall could swallow it entirely).
tracing::error!(error = %e, %client_udp, udp_port, "data-plane socket bind/hole-punch failed");
return Err(anyhow::Error::new(e)).context("bind data plane");
}
};
tracing::info!(
%client_udp,
punched,